Image inputs

Image inputs must first be uploaded and confirmed under the same account. External URLs, base64 and data URLs are not accepted. JPEG, PNG and WebP are supported; each file is at most 30 MiB and each preparation request contains at most five files.

1. Prepare uploads

curl --fail-with-body "$SEEDANCE_BASE_URL/uploads" \
  -H "Authorization: Bearer $SEEDANCE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"files":[{"name":"reference.png","type":"image/png","size":12345}]}'

Replace 12345 with the actual byte length. The response data[] contains asset_id, upload_url and upload_headers.

2. PUT the file bytes

PUT the exact file bytes to upload_url within five minutes, using every returned upload_headers entry, including Content-Length. Do not send your API Bearer key to the upload URL. Do not wrap the file in multipart/form-data.

3. Confirm the upload

curl --fail-with-body "$SEEDANCE_BASE_URL/uploads/complete" \
  -H "Authorization: Bearer $SEEDANCE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"asset_ids":["RETURNED_ASSET_ID"]}'

Use data[].url from the completion response as the generation input. Ownership, readiness, expiry and model-specific limits are checked before generation.

4. Generate with the image

{
  "model": "seedance-2.5-lite",
  "mode": "image-to-video",
  "prompt": "The subject slowly turns toward the camera.",
  "inputs": [
    {
      "type": "image",
      "url": "URL_FROM_UPLOAD_COMPLETE"
    }
  ],
  "parameters": {
    "duration_seconds": 8,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "image_mode": "first-frame"
  },
  "max_cost_credits": 80
}

Replace URL_FROM_UPLOAD_COMPLETE, then quote and POST this body to /videos with an idempotency key. For image-to-image use an image model, image-to-image, and /images, without video duration or image_mode.